/*
* Copyright 2010 The Closure Compiler Authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.google.javascript.jscomp;
import com.google.common.base.Joiner;
import com.google.common.base.Preconditions;
import com.google.common.collect.ImmutableSet;
import com.google.javascript.jscomp.CodingConvention.Bind;
import com.google.javascript.rhino.IR;
import com.google.javascript.rhino.Node;
import com.google.javascript.rhino.Token;
import java.util.regex.Pattern;
/**
* A peephole optimization that minimizes code by simplifying conditional
* expressions, replacing IFs with HOOKs, replacing object constructors
* with literals, and simplifying returns.
*
*/
class PeepholeSubstituteAlternateSyntax
extends AbstractPeepholeOptimization {
private static final CodeGenerator REGEXP_ESCAPER =
CodeGenerator.forCostEstimation(
null /* blow up if we try to produce code */);
private final boolean late;
private static final int STRING_SPLIT_OVERHEAD = ".split('.')".length();
static final DiagnosticType INVALID_REGULAR_EXPRESSION_FLAGS =
DiagnosticType.warning(
"JSC_INVALID_REGULAR_EXPRESSION_FLAGS",
"Invalid flags to RegExp constructor: {0}");
/**
* @param late When late is false, this mean we are currently running before
* most of the other optimizations. In this case we would avoid optimizations
* that would make the code harder to analyze (such as using string splitting,
* merging statements with commas, etc). When this is true, we would
* do anything to minimize for size.
*/
PeepholeSubstituteAlternateSyntax(boolean late) {
this.late = late;
}
/**
* Tries apply our various peephole minimizations on the passed in node.
*/
@Override
@SuppressWarnings("fallthrough")
public Node optimizeSubtree(Node node) {
switch(node.getType()) {
case Token.ASSIGN_SUB:
return reduceSubstractionAssignment(node);
case Token.TRUE:
case Token.FALSE:
return reduceTrueFalse(node);
case Token.NEW:
node = tryFoldStandardConstructors(node);
if (!node.isCall()) {
return node;
}
// Fall through on purpose because tryFoldStandardConstructors() may
// convert a NEW node into a CALL node
case Token.CALL:
Node result = tryFoldLiteralConstructor(node);
if (result == node) {
result = tryFoldSimpleFunctionCall(node);
if (result == node) {
result = tryFoldImmediateCallToBoundFunction(node);
}
}
return result;
case Token.RETURN:
return tryReduceReturn(node);
case Token.COMMA:
return trySplitComma(node);
case Token.NAME:
return tryReplaceUndefined(node);
case Token.ARRAYLIT:
return tryMinimizeArrayLiteral(node);
case Token.MUL:
case Token.AND:
case Token.OR:
case Token.BITOR:
case Token.BITXOR:
case Token.BITAND:
return tryRotateAssociativeOperator(node);
default:
return node; //Nothing changed
}
}
private Node tryRotateAssociativeOperator(Node n) {
if (!late) {
return n;
}
// All commutative operators are also associative
Preconditions.checkArgument(NodeUtil.isAssociative(n.getType()));
Node rhs = n.getLastChild();
if (n.getType() == rhs.getType()) {
// Transform a * (b * c) to a * b * c
Node first = n.getFirstChild().detachFromParent();
Node second = rhs.getFirstChild().detachFromParent();
Node third = rhs.getLastChild().detachFromParent();
Node newLhs = new Node(n.getType(), first, second)
.copyInformationFrom(n);
Node newRoot = new Node(rhs.getType(), newLhs, third)
.copyInformationFrom(rhs);
n.getParent().replaceChild(n, newRoot);
reportCodeChange();
return newRoot;
} else if (NodeUtil.isCommutative(n.getType()) &&
!NodeUtil.mayHaveSideEffects(n)) {
// Transform a * (b / c) to b / c * a
Node lhs = n.getFirstChild();
while (lhs.getType() == n.getType()) {
lhs = lhs.getFirstChild();
}
int precedence = NodeUtil.precedence(n.getType());
int lhsPrecedence = NodeUtil.precedence(lhs.getType());
int rhsPrecedence = NodeUtil.precedence(rhs.getType());
if (rhsPrecedence == precedence && lhsPrecedence != precedence) {
n.removeChild(rhs);
lhs.getParent().replaceChild(lhs, rhs);
n.addChildToBack(lhs);
reportCodeChange();
return n;
}
}
return n;
}
private Node tryFoldSimpleFunctionCall(Node n) {
Preconditions.checkState(n.isCall());
Node callTarget = n.getFirstChild();
if (callTarget != null && callTarget.isName() &&
callTarget.getString().equals("String")) {
// Fold String(a) to '' + (a) on immutable literals,
// which allows further optimizations
//
// We can't do this in the general case, because String(a) has
// slightly different semantics than '' + (a). See
// http://code.google.com/p/closure-compiler/issues/detail?id=759
Node value = callTarget.getNext();
if (value != null && value.getNext() == null &&
NodeUtil.isImmutableValue(value)) {
Node addition = IR.add(
IR.string("").srcref(callTarget),
value.detachFromParent());
n.getParent().replaceChild(n, addition);
reportCodeChange();
return addition;
}
}
return n;
}
private Node tryFoldImmediateCallToBoundFunction(Node n) {
// Rewriting "(fn.bind(a,b))()" to "fn.call(a,b)" makes it inlinable
Preconditions.checkState(n.isCall());
Node callTarget = n.getFirstChild();
Bind bind = getCodingConvention().describeFunctionBind(callTarget, false);
if (bind != null) {
// replace the call target
bind.target.detachFromParent();
n.replaceChild(callTarget, bind.target);
callTarget = bind.target;
// push the parameters
addParameterAfter(bind.parameters, callTarget);
// add the this value before the parameters if necessary
if (bind.thisValue != null && !NodeUtil.isUndefined(bind.thisValue)) {
// rewrite from "fn(a, b)" to "fn.call(thisValue, a, b)"
Node newCallTarget = IR.getprop(
callTarget.cloneTree(),
IR.string("call").srcref(callTarget));
n.replaceChild(callTarget, newCallTarget);
n.addChildAfter(bind.thisValue.cloneTree(), newCallTarget);
n.putBooleanProp(Node.FREE_CALL, false);
} else {
n.putBooleanProp(Node.FREE_CALL, true);
}
reportCodeChange();
}
return n;
}
private static void addParameterAfter(Node parameterList, Node after) {
if (parameterList != null) {
// push the last parameter to the head of the list first.
addParameterAfter(parameterList.getNext(), after);
after.getParent().addChildAfter(parameterList.cloneTree(), after);
}
}
private Node trySplitComma(Node n) {
if (late) {
return n;
}
Node parent = n.getParent();
Node left = n.getFirstChild();
Node right = n.getLastChild();
if (parent.isExprResult()
&& !parent.getParent().isLabel()) {
// split comma
n.detachChildren();
// Replace the original expression with the left operand.
parent.replaceChild(n, left);
// Add the right expression afterward.
Node newStatement = IR.exprResult(right);
newStatement.copyInformationFrom(n);
//This modifies outside the subtree, which is not
//desirable in a peephole optimization.
parent.getParent().addChildAfter(newStatement, parent);
reportCodeChange();
return left;
} else {
return n;
}
}
/**
* Use "void 0" in place of "undefined"
*/
private Node tryReplaceUndefined(Node n) {
// TODO(johnlenz): consider doing this as a normalization.
if (isASTNormalized()
&& NodeUtil.isUndefined(n)
&& !NodeUtil.isLValue(n)) {
Node replacement = NodeUtil.newUndefinedNode(n);
n.getParent().replaceChild(n, replacement);
reportCodeChange();
return replacement;
}
return n;
}
/**
* Reduce "return undefined" or "return void 0" to simply "return".
*
* @return The original node, maybe simplified.
*/
private Node tryReduceReturn(Node n) {
Node result = n.getFirstChild();
if (result != null) {
switch (result.getType()) {
case Token.VOID:
Node operand = result.getFirstChild();
if (!mayHaveSideEffects(operand)) {
n.removeFirstChild();
reportCodeChange();
}
break;
case Token.NAME:
String name = result.getString();
if (name.equals("undefined")) {
n.removeFirstChild();
reportCodeChange();
}
break;
}
}
return n;
}
private static final ImmutableSet<String> STANDARD_OBJECT_CONSTRUCTORS =
// String, Number, and Boolean functions return non-object types, whereas
// new String, new Number, and new Boolean return object types, so don't
// include them here.
ImmutableSet.of(
"Object",
"Array",
"Error"
);
/**
* Fold "new Object()" to "Object()".
*/
private Node tryFoldStandardConstructors(Node n) {
Preconditions.checkState(n.isNew());
if (canFoldStandardConstructors(n)) {
n.setType(Token.CALL);
n.putBooleanProp(Node.FREE_CALL, true);
reportCodeChange();
}
return n;
}
/**
* @return Whether "new Object()" can be folded to "Object()" on {@code n}.
*/
private boolean canFoldStandardConstructors(Node n) {
// If name normalization has been run then we know that
// new Object() does in fact refer to what we think it is
// and not some custom-defined Object().
if (isASTNormalized() && n.getFirstChild().isName()) {
String className = n.getFirstChild().getString();
if (STANDARD_OBJECT_CONSTRUCTORS.contains(className)) {
return true;
}
if ("RegExp".equals(className)) {
// Fold "new RegExp()" to "RegExp()", but only if the argument is a string.
// See issue 1260.
if (n.getChildAtIndex(1) == null || n.getChildAtIndex(1).isString()) {
return true;
}
}
}
return false;
}
/**
* Replaces a new Array, Object, or RegExp node with a literal, unless the
* call is to a local constructor function with the same name.
*/
private Node tryFoldLiteralConstructor(Node n) {
Preconditions.checkArgument(n.isCall()
|| n.isNew());
Node constructorNameNode = n.getFirstChild();
Node newLiteralNode = null;
// We require the AST to be normalized to ensure that, say,
// Object() really refers to the built-in Object constructor
// and not a user-defined constructor with the same name.
if (isASTNormalized() && Token.NAME == constructorNameNode.getType()) {
String className = constructorNameNode.getString();
if ("RegExp".equals(className)) {
// "RegExp("boo", "g")" --> /boo/g
return tryFoldRegularExpressionConstructor(n);
} else {
boolean constructorHasArgs = constructorNameNode.getNext() != null;
if ("Object".equals(className) && !constructorHasArgs) {
// "Object()" --> "{}"
newLiteralNode = IR.objectlit();
} else if ("Array".equals(className)) {
// "Array(arg0, arg1, ...)" --> "[arg0, arg1, ...]"
Node arg0 = constructorNameNode.getNext();
FoldArrayAction action = isSafeToFoldArrayConstructor(arg0);
if (action == FoldArrayAction.SAFE_TO_FOLD_WITH_ARGS ||
action == FoldArrayAction.SAFE_TO_FOLD_WITHOUT_ARGS) {
newLiteralNode = IR.arraylit();
n.removeChildren();
if (action == FoldArrayAction.SAFE_TO_FOLD_WITH_ARGS) {
newLiteralNode.addChildrenToFront(arg0);
}
}
}
if (newLiteralNode != null) {
n.getParent().replaceChild(n, newLiteralNode);
reportCodeChange();
return newLiteralNode;
}
}
}
return n;
}
private static enum FoldArrayAction {
NOT_SAFE_TO_FOLD, SAFE_TO_FOLD_WITH_ARGS, SAFE_TO_FOLD_WITHOUT_ARGS}
/**
* Checks if it is safe to fold Array() constructor into []. It can be
* obviously done, if the initial constructor has either no arguments or
* at least two. The remaining case may be unsafe since Array(number)
* actually reserves memory for an empty array which contains number elements.
*/
private static FoldArrayAction isSafeToFoldArrayConstructor(Node arg) {
FoldArrayAction action = FoldArrayAction.NOT_SAFE_TO_FOLD;
if (arg == null) {
action = FoldArrayAction.SAFE_TO_FOLD_WITHOUT_ARGS;
} else if (arg.getNext() != null) {
action = FoldArrayAction.SAFE_TO_FOLD_WITH_ARGS;
} else {
switch (arg.getType()) {
case Token.STRING:
// "Array('a')" --> "['a']"
action = FoldArrayAction.SAFE_TO_FOLD_WITH_ARGS;
break;
case Token.NUMBER:
// "Array(0)" --> "[]"
if (arg.getDouble() == 0) {
action = FoldArrayAction.SAFE_TO_FOLD_WITHOUT_ARGS;
}
break;
case Token.ARRAYLIT:
// "Array([args])" --> "[[args]]"
action = FoldArrayAction.SAFE_TO_FOLD_WITH_ARGS;
break;
default:
}
}
return action;
}
private Node tryFoldRegularExpressionConstructor(Node n) {
Node parent = n.getParent();
Node constructor = n.getFirstChild();
Node pattern = constructor.getNext(); // e.g. ^foobar$
Node flags = null != pattern ? pattern.getNext() : null; // e.g. gi
if (null == pattern || (null != flags && null != flags.getNext())) {
// too few or too many arguments
return n;
}
if (// is pattern folded
pattern.isString()
// make sure empty pattern doesn't fold to //
&& !"".equals(pattern.getString())
&& (null == flags || flags.isString())
// don't escape patterns with Unicode escapes since Safari behaves badly
// (read can't parse or crashes) on regex literals with Unicode escapes
&& (isEcmaScript5OrGreater()
|| !containsUnicodeEscape(pattern.getString()))) {
// Make sure that / is escaped, so that it will fit safely in /brackets/
// and make sure that no LineTerminatorCharacters appear literally inside
// the pattern.
// pattern is a string value with \\ and similar already escaped
pattern = makeForwardSlashBracketSafe(pattern);
Node regexLiteral;
if (null == flags || "".equals(flags.getString())) {
// fold to /foobar/
regexLiteral = IR.regexp(pattern);
} else {
// fold to /foobar/gi
if (!areValidRegexpFlags(flags.getString())) {
report(INVALID_REGULAR_EXPRESSION_FLAGS, flags);
return n;
}
if (!areSafeFlagsToFold(flags.getString())) {
return n;
}
n.removeChild(flags);
regexLiteral = IR.regexp(pattern, flags);
}
parent.replaceChild(n, regexLiteral);
reportCodeChange();
return regexLiteral;
}
return n;
}
private Node reduceSubstractionAssignment(Node n) {
Node right = n.getLastChild();
if (right.isNumber()) {
if (right.getDouble() == 1) {
Node newNode = IR.dec(n.removeFirstChild(), false);
n.getParent().replaceChild(n, newNode);
reportCodeChange();
return newNode;
} else if (right.getDouble() == -1) {
Node newNode = IR.inc(n.removeFirstChild(), false);
n.getParent().replaceChild(n, newNode);
reportCodeChange();
return newNode;
}
}
return n;
}
private Node reduceTrueFalse(Node n) {
if (late) {
switch (n.getParent().getType()) {
case Token.EQ:
case Token.GT:
case Token.GE:
case Token.LE:
case Token.LT:
case Token.NE:
Node number = IR.number(n.isTrue() ? 1 : 0);
n.getParent().replaceChild(n, number);
reportCodeChange();
return number;
}
Node not = IR.not(IR.number(n.isTrue() ? 0 : 1));
not.copyInformationFromForTree(n);
n.getParent().replaceChild(n, not);
reportCodeChange();
return not;
}
return n;
}
private Node tryMinimizeArrayLiteral(Node n) {
boolean allStrings = true;
for (Node cur = n.getFirstChild(); cur != null; cur = cur.getNext()) {
if (!cur.isString()) {
allStrings = false;
}
}
if (allStrings) {
return tryMinimizeStringArrayLiteral(n);
} else {
return n;
}
}
private Node tryMinimizeStringArrayLiteral(Node n) {
if (!late) {
return n;
}
int numElements = n.getChildCount();
// We save two bytes per element.
int saving = numElements * 2 - STRING_SPLIT_OVERHEAD;
if (saving <= 0) {
return n;
}
String[] strings = new String[n.getChildCount()];
int idx = 0;
for (Node cur = n.getFirstChild(); cur != null; cur = cur.getNext()) {
strings[idx++] = cur.getString();
}
// These delimiters are chars that appears a lot in the program therefore
// probably have a small Huffman encoding.
String delimiter = pickDelimiter(strings);
if (delimiter != null) {
String template = Joiner.on(delimiter).join(strings);
Node call = IR.call(
IR.getprop(
IR.string(template),
IR.string("split")),
IR.string("" + delimiter));
call.copyInformationFromForTree(n);
n.getParent().replaceChild(n, call);
reportCodeChange();
return call;
}
return n;
}
/**
* Find a delimiter that does not occur in the given strings
* @param strings The strings that must be separated.
* @return a delimiter string or null
*/
private static String pickDelimiter(String[] strings) {
boolean allLength1 = true;
for (String s : strings) {
if (s.length() != 1) {
allLength1 = false;
break;
}
}
if (allLength1) {
return "";
}
String[] delimiters = new String[]{" ", ";", ",", "{", "}", null};
int i = 0;
NEXT_DELIMITER: for (; delimiters[i] != null; i++) {
for (String cur : strings) {
if (cur.contains(delimiters[i])) {
continue NEXT_DELIMITER;
}
}
break;
}
return delimiters[i];
}
private static final Pattern REGEXP_FLAGS_RE = Pattern.compile("^[gmi]*$");
/**
* are the given flags valid regular expression flags?
* JavaScript recognizes several suffix flags for regular expressions,
* 'g' - global replace, 'i' - case insensitive, 'm' - multi-line.
* They are case insensitive, and JavaScript does not recognize the extended
* syntax mode, single-line mode, or expression replacement mode from Perl 5.
*/
private static boolean areValidRegexpFlags(String flags) {
return REGEXP_FLAGS_RE.matcher(flags).matches();
}
/**
* are the given flags safe to fold?
* We don't fold the regular expression if global ('g') flag is on,
* because in this case it isn't really a constant: its 'lastIndex'
* property contains the state of last execution, so replacing
* 'new RegExp('foobar','g')' with '/foobar/g' may change the behavior of
* the program if the RegExp is used inside a loop, for example.
* <p>
* ECMAScript 5 explicitly disallows pooling of regular expression literals so
* in ECMAScript 5, {@code /foo/g} and {@code new RegExp('foo', 'g')} are
* equivalent.
* From section 7.8.5:
* "Then each time the literal is evaluated, a new object is created as if by
* the expression new RegExp(Pattern, Flags) where RegExp is the standard
* built-in constructor with that name."
*/
private boolean areSafeFlagsToFold(String flags) {
return isEcmaScript5OrGreater() || flags.indexOf('g') < 0;
}
/**
* returns a string node that can safely be rendered inside /brackets/.
*/
private static Node makeForwardSlashBracketSafe(Node n) {
String s = n.getString();
// sb contains everything in s[0:pos]
StringBuilder sb = null;
int pos = 0;
boolean isEscaped = false, inCharset = false;
for (int i = 0; i < s.length(); ++i) {
char ch = s.charAt(i);
switch (ch) {
case '\\':
isEscaped = !isEscaped;
continue;
case '/':
// Escape a literal forward slash if it is not already escaped and is
// not inside a character set.
// new RegExp('/') -> /\//
// but the following do not need extra escaping
// new RegExp('\\/') -> /\//
// new RegExp('[/]') -> /[/]/
if (!isEscaped && !inCharset) {
if (null == sb) { sb = new StringBuilder(s.length() + 16); }
sb.append(s, pos, i).append('\\');
pos = i;
}
break;
case '[':
if (!isEscaped) {
inCharset = true;
}
break;
case ']':
if (!isEscaped) {
inCharset = false;
}
break;
case '\r': case '\n': case '\u2028': case '\u2029':
// LineTerminators cannot appear raw inside a regular
// expression literal.
// They can't appear legally in a quoted string, but when
// the quoted string from
// new RegExp('\n')
// reaches here, the quoting has been removed.
// Requote just these code-points.
if (null == sb) { sb = new StringBuilder(s.length() + 16); }
if (isEscaped) {
sb.append(s, pos, i - 1);
} else {
sb.append(s, pos, i);
}
switch (ch) {
case '\r': sb.append("\\r"); break;
case '\n': sb.append("\\n"); break;
case '\u2028': sb.append("\\u2028"); break;
case '\u2029': sb.append("\\u2029"); break;
}
pos = i + 1;
break;
}
isEscaped = false;
}
if (null == sb) { return n.cloneTree(); }
sb.append(s, pos, s.length());
return IR.string(sb.toString()).srcref(n);
}
/**
* true if the JavaScript string would contain a Unicode escape when written
* out as the body of a regular expression literal.
*/
static boolean containsUnicodeEscape(String s) {
String esc = REGEXP_ESCAPER.regexpEscape(s);
for (int i = -1; (i = esc.indexOf("\\u", i + 1)) >= 0;) {
int nSlashes = 0;
while (i - nSlashes > 0 && '\\' == esc.charAt(i - nSlashes - 1)) {
++nSlashes;
}
// if there are an even number of slashes before the \ u then it is a
// Unicode literal.
if (0 == (nSlashes & 1)) { return true; }
}
return false;
}
}